Plant diagnostics by transient classification: The ALADDIN approach

Abstract

Any action taken on a plant, for example in response to an abnormal situation or in reaction to unsafe conditions, relies on the ability to identify the state and dynamics of operation of the plant. Although there might be hundreds or even thousands of measurements in a plant, there are generally few events occurring. The data from these measurements must be mapped into appropriate descriptions of the occurring event(s), which in most cases is a difficult task. The real-time history of scores of variables can be displayed and monitored in most computerized plant monitoring and control systems. However, whereas a simple visual inspection of displayed trends is generally sufficient to allow the operator to confirm the plant status during normal, steady-state operations, when the plant is subject to deviations due to anomalies or faults, the displayed trends of interacting variables can be very difficult to interpret, either because the changes are too subtle, or because the changes are too fast. In this article we describe the ALADDIN methodology for dynamic event recognition and fault diagnosis, which combines techniques such as recurrent neural network ensembles, wavelet on-line pre-processing (WOLP), and autonomous recursive task decomposition (ARTD), in an attempt to improve the practical applicability and scalability of this type of system to real processes and machinery. © 2002 Wiley Periodicals, Inc.